Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: Ali Chen who wrote (156292)	5/20/2005 12:21:44 AM
From: THE WATSONYOUTH	Read Replies (2) \| Respond to of 275872

I believe the reason for this across-die variability is in vastly emphasised statistical nature of thermal processes involved at deep submicron semi-quantum level, which I have mentioned on this thead many times before. , Hi Ali, I saw your discussion with Elmo on the Intel thread and chose to respond here. The across die variability you describe is NOT primarily due to the statistical nature of thermal processes involved at deep submicron semi quantum level. The overwhelming cause of across die variability is ACLV (across chip linewidth variation) The reason it gets worse with each generation is the fact that channel length tolerances have not scaled with each generation as they must. Even for virtual ideal processing (Lith/RIE/thermal/implant/etc.), channel length variations due to physical mask variations can limit the number of the fastest parts as the physical design limit of the device is approached. What becomes difficult is determining the relationship of things you can easily measure in the kerf with things you can't measure in the product. For instance, you can measure things in the kerf like ring oscillator performance, channel lengths via gate capacitance measurements of long chain structures, ACLV via poly resistance of specially designed structures, etc. But mask variations may be such that the physical poly length of these structures are different than say the ave physical poly length in the product and perhaps very different than the length of the physical poly in the devices in a critical path which may determine max freq. Even if the mask produces a distribution which meets the ACLV target (say +/-50A 3 sigma for a 90nm process), the kerf measurements may all come from near the long end of the distribution while the actual poly length in the critical path that may determine max freq may be from the shortest part of the distribution. Or.....perhaps the relationship may be reversed. The bottom line is that two supposed identical masks may have very different relationships between what can be routinely measured in the kerf and the final max freq of the product. In the case of small die where multiple chips are shot per exposure field, the chips will usually show a very distinct speed distribution corresponding to their field position at gate exposure. This will match the poly linewidth distribution on the mask. So, the largest part of within chip variation is in the making of the masks. I think the reduction on exposure is only 4x or 5x....so you are all ready dealing with sub .50u chrome. Think about it. THE WATSONYOUTH